从 R 中的文本中提取字符级 n-gram

Question

我有一个带有文本的数据框，我想提取字符级双字母组 (n = 2)，例如"st"、"ac"、"ck"，对于 R 中的每个文本。

我也想统计每个字符级别的二元组在文本中出现的频率

数据：

df$text

[1] "hy my name is"
[2] "stackover flow is great"
[3] "how are you"

Answer 1

我不太确定您在这里的预期输出。我原以为 "stack" 的二元组会是 "st"、"ta"、"ac" 和 "ck"，因为这会捕获每个连续的对。

例如，如果你想知道单词 "brothers" 中有多少个二元组 "th" 的实例，你将其拆分为二元组 "br"、"ot"、"he" 和 "rs"，那么你会得到答案 0，这是错误的。

您可以构建一个函数来获取所有个二元语法，如下所示：

# This function takes a vector of single characters and creates all the bigrams
# within that vector. For example "s", "t", "a", "c", "k" becomes 
# "st", "ta", "ac", and "ck"

pair_chars <- function(char_vec) {
  all_pairs <- paste0(char_vec[-length(char_vec)], char_vec[-1])
  return(as.vector(all_pairs[nchar(all_pairs) == 2]))
}

# This function splits a single word into a character vector and gets its bigrams

word_bigrams <- function(words){
  unlist(lapply(strsplit(words, ""), pair_chars))
}

# This function splits a string or vector of strings into words and gets their bigrams

string_bigrams <- function(strings){
  unlist(lapply(strsplit(strings, " "), word_bigrams))
}

所以现在我们可以在您的示例上进行测试：

df <- data.frame(text = c("hy my name is", "stackover flow is great", 
                          "how are you"), stringsAsFactors = FALSE)

string_bigrams(df$text)
#>  [1] "hy" "my" "na" "am" "me" "is" "st" "ta" "ac" "ck" "ko" "ov" "ve" "er" "fl"
#> [16] "lo" "ow" "is" "gr" "re" "ea" "at" "ho" "ow" "ar" "re" "yo" "ou"

如果你想计算出现次数，你可以使用table:

table(string_bigrams(df$text))

#> ac am ar at ck ea er fl gr ho hy is ko lo me my na ou ov ow re st ta ve yo 
#>  1  1  1  1  1  1  1  1  1  1  1  2  1  1  1  1  1  1  1  2  2  1  1  1  1

但是，如果您要进行大量的文本挖掘，您应该查看特定的 R 包，例如 stringi、stringr、tm 和 quanteda 有助于完成基本任务

例如，我上面写的所有基本 R 函数都可以使用 quanteda 包替换，如下所示：

library(quanteda)
char_ngrams(unlist(tokens(df$text, "character")), concatenator = "")
#>  [1] "hy" "ym" "my" "yn" "na" "am" "me" "ei" "is" "ss" "st" "ta" "ac" "ck" 
#> [15] "ko" "ov" "ve" "er" "rf" "fl" "lo" "ow" "wi" "is" "sg" "gr" "re" "ea"
#> [29] "at" "th" "ho" "ow" "wa" "ar" "re" "ey" "yo" "ou"

^{由 reprex package (v0.3.0)}

于 2020-06-13 创建

Answer 2

除了艾伦的回答，

您可以结合使用 stringdist 包中的 qgram 函数和 gsub 来删除空格。

library(stringdist)
qgrams(gsub(" ", "", df1$text), q = 2)

   hy ym yn yo my na st ta ve wi wa ov rf sg ow re ou me is ko lo am ei er fl gr ho ey ck ea at ar ac
V1  1  1  1  1  1  1  1  1  1  1  1  1  1  1  2  2  1  1  2  1  1  1  1  1  1  1  1  1  1  1  1  1  1

从 R 中的文本中提取字符级 n-gram

Extract character-level n-grams from text in R

r

nlp

character

n-gram